Hard Materials Seminar - "Mechanism-Driven AI for Alloy Design: From Uncertainty-aware Physics-based ML Prediction of Plasticity to LLM-guided Symbolic Regression"

Designing alloys for extreme environments requires reliable predictions of mechanical behavior and an understanding of the mechanisms governing deformation. While artificial intelligence (AI) offers new opportunities to accelerate alloy design, materials data are often sparse and heterogeneous, and the relevant physical mechanisms remain incompletely understood. In this talk, I will present a mechanism-driven approach that connects probabilistic machine learning, symbolic regression, and multiscale simulations to address these challenges. 

I will first introduce physics-based, uncertainty-aware machine learning models that describe dislocation density evolution and predict the stress-strain response of polycrystalline metals. By combining learned evolution laws with theoretical models of strengthening, this framework extends predictions from pure metals to FCC and BCC multi-principal element alloys while quantifying uncertainty. Next, I will discuss LLM-guided symbolic regression to identify interpretable equations of dislocation density evolution, with physical knowledge guiding the search for expressions that generalize across materials. Finally, I will present a machine learning-bridged multiscale framework linking molecular dynamics and discrete dislocation dynamics to investigate how chemical short-range order influences collective dislocation behavior. 

These studies establish a foundation for integrating mechanism-informed prediction with physical mechanism discovery. I will conclude with a vision for trustworthy AI-enabled mechanism-driven closed-loop alloy design, in which simulations, experiments, and knowledge infrastructure jointly guide the discovery of mechanisms and the design of new alloys.